Petroleum chemical abnormality detection method and system based on multi-modal attention fusion

By employing a multimodal attention fusion method, combined with feature extraction and classification of visual and UWB data, the real-time and accuracy issues of anomaly detection in petrochemical production sites were resolved, achieving high-precision, low-false-report multi-level alarms.

CN121884065BActive Publication Date: 2026-06-19CHENGDU GREATECH ELECTRONIC TECHNOLOGY CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CHENGDU GREATECH ELECTRONIC TECHNOLOGY CO LTD
Filing Date
2026-03-18
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing petrochemical production site monitoring systems struggle to fully leverage the complementary features of vision and UWB under complex operating conditions, resulting in slow anomaly detection response, low accuracy, high false alarm rate, and a lack of real-time and hierarchical alarm functions.

Method used

A multimodal attention fusion method is adopted, which extracts features from video frames and UWB positioning data through convolutional neural networks, and combines them with a multi-head self-attention classification network for feature mapping and weighting processing to achieve bidirectional compensation and deep coupling of information, generating high-precision and low-false-alarm-rate abnormal alarms.

Benefits of technology

It achieves high-precision anomaly detection under complex working conditions, reduces false alarm rate, and can generate multi-level alarms in real time, making up for the shortcomings of traditional systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121884065B_ABST
    Figure CN121884065B_ABST
Patent Text Reader

Abstract

This invention relates to a method and system for anomaly detection in petrochemical plants based on multimodal attention fusion. The method includes: acquiring real-time video frames from a petrochemical plant and preprocessing the frames to obtain binary images; using a convolutional neural network to extract original features from the binary images, and mapping and weighting these features using attention weights to obtain fused features; linearly projecting the fused features to calculate attention outputs based on the projection matrix of each head, and concatenating and linearly mapping the attention outputs of all heads; performing residual connections and normalization on the fused features based on the concatenation and linear mapping of the attention outputs of all heads to obtain a normalized result; and inputting the normalized result into a classification head to output an anomaly alarm for the corresponding class. This invention, combined with a multi-head self-attention classification network, improves the accuracy of anomaly detection, reduces the false alarm rate, and provides high real-time alarm performance.
Need to check novelty before this filing date? Find Prior Art